CN108510499A - A kind of carrying out image threshold segmentation method and device based on fuzzy set and Otsu - Google Patents

Abstract

The present invention relates to image segmentation fields, and in particular to a kind of carrying out image threshold segmentation method and device based on fuzzy set and Otsu.The present invention is primarily based on fuzzy set and gives new enhanced fuzzy membership function；Then the discretization method for utilizing mean square deviation, constructs the inter-class variance of Otsu；Renyi entropy theories are finally combined, the Renyi entropys that weight calculation provides image are introduced, the threshold value for reusing maximum Renyi entropys completes image segmentation.The present invention compares traditional Threshold Segmentation Algorithm, has preferable advantage, while also with good stability and segmentation effect in the accuracy in segmenting edge and the robustness to noise, can effectively improve the precision of image segmentation.

Description

A Method and Device for Image Threshold Segmentation Based on Fuzzy Sets and Otsu

technical field

The invention relates to the field of image segmentation, in particular to an image threshold segmentation method and device based on fuzzy sets and Otsu.

Background technique

Image segmentation generally refers to dividing the image into several non-overlapping target and background regions according to the features such as grayscale, color, texture, shape and edge in the image, and making these features show similarity in the same region. , and there are obvious differences in different regions.

Among many image segmentation methods, the threshold segmentation method has become the most widely used segmentation technology in image segmentation because of its simplicity, effectiveness, low computational complexity, and stable performance. The key is how to choose the threshold in order to obtain the optimal segmentation effect. Threshold segmentation methods can be roughly divided into two categories: global threshold segmentation methods and local threshold segmentation methods. The global threshold segmentation method selects a single threshold based on the entire image histogram information to divide the image into two parts; the local threshold segmentation method divides the original image into multiple smaller images, and selects the corresponding threshold for each sub-image.

A large number of threshold selection methods have been proposed in the prior art, combined with intelligent algorithms to find the optimal threshold under different criteria, and achieved good application results in different application fields. Sezgin and Sankur wrote "Surveyover image thresholding techniques and quantitative performance evaluation" (Journal of Electronic Imaging, 2004, 13(1): 146-166.) in 2004, and made a comprehensive comparison of more than 40 global threshold segmentation methods. It is pointed out that threshold segmentation methods based on Otsu and maximum entropy are two widely used global threshold segmentation methods. The Otsu algorithm is based on the statistical characteristics of the whole image, and realizes the automatic selection of the image threshold. The segmentation effect is good, and it has been widely used in practice. As far as Otsu is concerned, the advantage is that the algorithm is simple, and when the area of the target and the background is not much different, the image can be segmented very effectively. However, this algorithm has a large amount of calculation and is difficult to adapt to real-time processing; at the same time, when the area of the target and the background in the image is very different, the segmentation effect is not good, showing that the histogram has no obvious double peaks or two peaks The size is very different, and when the grayscale of the target and the background has a large overlap, the two cannot be separated accurately. The reason for this phenomenon is that on the one hand, the method ignores the spatial information of the image, and on the other hand, the gray distribution of the image is used as the basis for segmenting the image, so it is also quite sensitive to noise. Thus, in practice, Otsu is always used in combination with other methods.

Contents of the invention

The object of the present invention is to provide a method and device for image threshold segmentation based on fuzzy sets and Otsu to solve the problem of poor effect of image segmentation in existing methods.

To achieve the above object, the invention provides a method for image threshold segmentation based on fuzzy sets and Otsu, including:

Method scheme one includes the following steps:

Use the fuzzy algorithm to perform fuzzy enhancement processing on the original image;

Normalize the enhanced new image;

Calculate the threshold th1 of Renyi entropy and the threshold th2 of Otsu respectively;

Respectively find the inter-class variance corresponding to the threshold th1 and the inter-class variance corresponding to the threshold th2;

Calculate the weight S 1 of the inter-class variance and the weight S 2 corresponding to the Renyi entropy according to the inter-class variance corresponding to the threshold _th1 and the threshold _th2 ;

Obtain the segmentation threshold th=S ₁ th2+S ₂ th1; perform image segmentation according to the segmentation threshold th.

Method 2: On the basis of method 1, the formulas for calculating Renyi’s entropy threshold th1 and Otsu’s threshold th2 include:

Among them, th1 is the threshold of Renyi entropy; th2 is the threshold of Otsu; E _O is the Renyi entropy of the target domain of the original image; E _B is the Renyi entropy of the background domain of the original image; L is the gray level of the original image; t is the gray level Threshold, and the value of t is [0, L–1]; w ₀ is the proportion of the first gray scale, w ₁ is the proportion of the second gray scale, the first gray scale and the second gray scale The grayscale is divided according to the grayscale threshold t; _σ0 is the grayscale mean square error of the image corresponding to the first type of grayscale, and _σ1 is the grayscale mean square error of the image corresponding to the second type of grayscale; The average gray value after the blur enhancement and the normalization processing.

Method scheme three, on the basis of method scheme one or method scheme two, the process of calculating the variance between classes includes:

The between-class variances _σ2 and _σ3 of the threshold th1 of Renyi entropy are:

The between-class variances _σ0 and _σ1 of Otsu's threshold th2 are:

Among them, p _i is the probability of occurrence of the gray level with the number of pixels n _i ; u _T is the average gray level of the original image, L1=0, L2=L–1.

Method scheme four, on the basis of method scheme three, the process of calculating the weight of the inter-class variance includes:

Wherein, S ₁ is the weight of the inter-class variance, and S ₂ =1−S ₁ .

Method scheme five, on the basis of method scheme four, the normalization process includes:

Use min or max operator for edge extraction;

Truncating the extracted edge data;

The truncation process is:

Wherein, Tr(u _ij ) is the edge data; u _ij is the membership function in the fuzzy algorithm.

The present invention also provides an image threshold segmentation device based on fuzzy sets and Otsu, comprising:

The device solution one includes a processor and a memory, and the processor stores instructions for implementing the following method:

Use the fuzzy algorithm to perform fuzzy enhancement processing on the original image;

Normalize the enhanced new image;

Calculate the threshold th1 of Renyi entropy and the threshold th2 of Otsu respectively;

Respectively find the inter-class variance corresponding to the threshold th1 and the inter-class variance corresponding to the threshold th2;

Calculate the weight S 1 of the inter-class variance and the weight S 2 corresponding to the Renyi entropy according to the inter-class variance corresponding to the threshold _th1 and the threshold _th2 ;

Obtain the segmentation threshold th=S ₁ th2+S ₂ th1; perform image segmentation according to the segmentation threshold th.

Device scheme 2, on the basis of device scheme 1, the formulas for calculating the threshold th1 of Renyi entropy and the threshold th2 of Otsu include:

Among them, th1 is the threshold of Renyi entropy; th2 is the threshold of Otsu; E _O is the Renyi entropy of the target domain of the original image; E _B is the Renyi entropy of the background domain of the original image; L is the gray level of the original image; t is the gray level Threshold, and the value of t is [0, L–1]; w ₀ is the proportion of the first gray scale, w ₁ is the proportion of the second gray scale, the first gray scale and the second gray scale The grayscale is divided according to the grayscale threshold t; _σ0 is the grayscale mean square error of the image corresponding to the first type of grayscale, and _σ1 is the grayscale mean square error of the image corresponding to the second type of grayscale; The average gray value after the blur enhancement and the normalization processing.

Device scheme three, on the basis of device scheme one or device scheme two, the process of calculating the variance between the classes includes:

The between-class variances _σ2 and _σ3 of the threshold th1 of Renyi entropy are:

The between-class variances _σ0 and _σ1 of Otsu's threshold th2 are:

Among them, p _i is the probability of occurrence of the gray level with the number of pixels n _i ; u _T is the average gray level of the original image, L1=0, L2=L–1.

Device scheme four, on the basis of device scheme three, the process of calculating the weight of the variance between classes includes:

Wherein, S ₁ is the weight of the inter-class variance, and S ₂ =1−S ₁ .

Device scheme five, on the basis of device scheme four, the normalization process includes:

Use min or max operator for edge extraction;

Truncating the extracted edge data;

The truncation process is:

Wherein, Tr(u _ij ) is the edge data; u _ij is the membership function in the fuzzy algorithm.

The beneficial effects of the present invention are: firstly, a new fuzzy enhanced membership function is given based on the fuzzy set; then, the discretization method of the mean square error is used to construct the inter-class variance of Otsu; finally, combined with the Renyi entropy theory, the weight calculation is introduced to give the image Renyi entropy, and then use the threshold of the maximum Renyi entropy to complete image segmentation. Compared with the traditional threshold segmentation algorithm, the present invention has better advantages in the accuracy of edge segmentation and robustness to noise, and also has good stability and segmentation effect, and can effectively improve the accuracy of image segmentation.

Description of drawings

Fig. 1 is a flow chart of the inventive method;

Fig. 2 is an effect diagram of one iteration of the Lena image using the Pal-King method;

Fig. 3 is the rendering of the second iteration of the Lena image using the Pal-King method;

Fig. 4 is the effect figure that adopts the method of the present invention to iterate once to Lena image;

Figure 5 is a Cameraman image;

Figure 6 is a Cameraman blur enhanced image;

Figure 7 is the segmentation result of the Cameraman image using the traditional Otsu algorithm;

Fig. 8 adopts the method of the present invention to the segmentation result of Cameraman image;

Figure 9 is a histogram of the Cameraman image;

Figure 10 is a Goldhill image;

Figure 11 is a Goldhill fuzzy enhanced image;

Fig. 12 is the segmentation result of the Goldhill image using the traditional Otsu algorithm;

Fig. 13 adopts the method of the present invention to the segmentation result of Goldhill image;

Figure 14 is a histogram of the Goldhill image.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

In order to solve the problem that the traditional Otsu algorithm has inconspicuous and inaccurate segmentation effects on images containing noise and uneven illumination, an image threshold segmentation method and device based on fuzzy sets and Otsu is proposed. Firstly, a new fuzzy enhanced membership function is given based on fuzzy sets; then, Otsu’s inter-class variance is constructed using the discretization method of mean square error; finally, combined with Renyi entropy theory, weight calculation is introduced to give the Renyi entropy of the image, and then Image segmentation is done using a threshold of the maximum Renyi entropy.

According to the concept of fuzzy sets, an image with a size of M×N and a gray level of L can be expressed as an M×N fuzzy matrix as follows:

Each element in the matrix Indicates the membership degree of the gray level x _ij of the pixel (i, j) in the image relative to the corresponding gray level x in the original image. This is a problem of finding a fuzzy distribution. Among the classic fuzzy enhancement algorithms, Pal and King proposed a fuzzy enhancement algorithm in which the membership function used is:

Among them, the parameters F _d and F _e are related to the shape of u _ij and can be determined by the transition point. In general, after taking F _e = 2 to obtain u _ij , the image needs to be blurred and enhanced, and the following transformation is adopted:

u _ij ＝T _r (u _ij )＝T ₁ (T _r-1 (u _ij )) (3)

where r = 1, 2, . . .

When u _ij >0.5, increase the value of u _ij ; when u _ij ≤0.5, decrease the value of u _ij . Perform inverse transformation on u _ij , after blur enhancement, get the gray value of the pixel (i, j) in the image X', X':

x _ij ＝T ^-1 (u _ij ) (5)

Among them, T ^-1 (·) is the inverse operation of T(·) in formula (2).

The Ostu method is based on the gray histogram of the image, and the maximum variance between the classes of the target and the background is used as the threshold selection criterion. The basic idea: assuming that the original grayscale image has L grayscale levels in total, the number of pixels of a certain grayscale level is n _i , and the total pixels of the image are N, then the probability of occurrence of each grayscale level can be obtained:

In image segmentation, according to the gray level of the image, the gray level is divided into two categories with the threshold t, namely C ₀ class (gray level is 0, 1, 2,..., t) and C ₁ class (gray level is t +1,t+2,...,L–1), the range of t is determined by normalization processing, and the initial value of t is the minimum value corresponding to the presence of gray levels (the probability of gray level occurrence is not equal to 0). value. The proportions of C ₀ and C ₁ appearing are:

Therefore, the C ₀ mean and C ₁ mean are respectively:

The average gray value of the entire image is:

So, the between-class variance is:

σ _B ² ＝ω ₀ (u ₀ -u _T ) ² +ω ₁ (u ₁ -u _T ) ² (12)

Let t take a value between [0, L–1], and the threshold that maximizes the distance between the two parts of the image is the optimal threshold of the Otsu method, and its expression:

1. Improved blur enhancement

Suppose an image is X, its size is M×N, the gray level is L, the maximum gray level is L–1, x _ij represents the gray value of the (i, j)th pixel of the image, then the new membership A function can be defined as:

The image is enhanced using repeated recursive calls as follows:

u _ij =T _r (u _ij )=T _r (T _r-1 (u _ij )) (15)

where r=1,2,...,∞.

After multiple regression calls, the operator T _r (u _ij ) enhances the larger membership degree value and suppresses the smaller membership degree value.

Normalized processing. Use the "min" or "max" operator for edge extraction, and truncate the extracted "edge" data T _r (u _ij ):

After the truncation processing of formula (17), the image data can be converted from the fuzzy domain to the spatial domain of the image, that is, the grayscale domain of the image.

In order to verify the effectiveness of the improved fuzzy enhancement on image segmentation, the Lena image is selected and enhanced by the Pal-King method in the traditional fuzzy enhancement and the improved fuzzy enhancement method of the present invention. The simulation experiment results are shown in Fig. 2-Fig. 4. The image enhancement effect of the method of the present invention in FIG. 4 after one iteration is better than that of the Pal-King method in FIG. 3 after two iterations. The improved fuzzy enhancement process preserves the edge information of low gray value in the image, and then preserves the integrity of image information, which is helpful for the next step of image segmentation.

2. Renyi entropy

Entropy is the basic method to describe uncertain factors in information theory, and the boundary distribution of images is the most uncertain. Therefore, the entropy (the largest amount of information) at the junction of the target and the background in the image is the largest, and the entropy reflects the overall outline of the image. The basic concept of Renyi entropy is given below: Let the probabilities of target O and target B be respectively:

And there is P _O (t)+P _B (t)=1. Therefore, the Renyi entropy corresponding to the image target domain and background domain can be defined as:

Then the overall Renyi entropy of the image is defined as:

Where α>0, α is a parameter, which is set to 0.7 in this paper.

According to the threshold selection principle of maximum entropy image segmentation, a certain threshold t can make formula (22) obtain the maximum value, then it is the optimal segmentation threshold, namely:

3. Improved Otsu model

Normally, different objects have relatively uniform internal gray levels, and the gray levels of pixels distributed at the junction between objects and their vicinity generally vary greatly, and the mean square error value can reflect the degree of dispersion of gray levels, that is, the gray level distribution uniformity. Therefore, the gray level change of the boundary can be approximately reflected by the mean square error value of the image.

The gray mean square error σ ₀ and σ ₁ of the given images C ₀ and C ₁ are expressed as:

The average gray value of the entire image after processing is expressed as:

The between-class variance is expressed as:

σ _B1 ² ＝w ₀ (σ ₀ -σ) ² +w ₁ (σ ₁ -σ) ² (27)

Bring the threshold t calculated by the Renyi entropy method into formulas (24) and (25), and calculate its variance as σ ₂ ² and σ ₃ ² , respectively, and the image can be divided into two targets:

Among them, L1=0, L2=L-1.

The weight corresponding to the threshold calculated by the maximum between-class variance method is expressed as:

Then the weight corresponding to the threshold calculated by Renyi entropy is 1–S.

The inventive method combines Renyi entropy and improved Otsu model, has proposed the image threshold segmentation algorithm (FSO-ITS) based on fuzzy set and Otsu, as shown in Figure 1, its specific steps are described as follows:

Step 1: Input the original image;

Step 2: use the new membership function to perform fuzzy enhancement processing on the original image;

Step 3: normalize the enhanced image;

Step 4: Calculate Renyi entropy and improved Otsu thresholds th1 and th2 respectively;

Step 5: Calculate the inter-class variance σ ₂ and σ ₃ of th1 and the inter-class variance σ ₀ and σ ₁ of th2 respectively;

Step 6: Use the ratio of the inter-class variance of th2 to the total inter-class variance as the weight S;

Step 7: Use weights to calculate the threshold th when the sum of image Renyi entropy and inter-class variance reaches the maximum, and finally segment the image.

In the above steps, the formula is included:

4. Experimental results and analysis

Experimental platform: Pentium4CPU3.0GHz dual-core PC with 2GB memory, Windows7 operating system and Matlab7.0 operating environment.

The traditional Otsu algorithm (Otsu N.A threshold selection method from gray-level histograms [J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1): 62-66.), Renyi entropy algorithm (Jizba P, Arimitsu T .Observability of Renyi's entropy[J].Physical Review E,2004,69(2):026128.) and the FSO-ITS algorithm of the present invention are compared and analyzed experimentally.

In order to analyze and verify the actual effect of the proposed improved algorithm, a standard picture Cameraman with 256×256 pixels was selected in the experiment, as shown in Figure 5. Firstly, blur enhancement processing is performed on the original image Figure 5, and the result is shown in Figure 6. Then, the traditional Otsu algorithm and the FSO-ITS algorithm are used to segment the processed image, and the segmentation results are shown in Figure 7 and Figure 8, respectively. Figure 9 is a histogram of the original image.

It can be seen intuitively from Figure 7 and Figure 8 that compared with the traditional Otsu algorithm, the image obtained by FSO-ITS algorithm segmentation has better denoising effect and clearer edges. In order to further verify the effectiveness of the FSO-ITS algorithm, another standard image Goldhill with 512 pixels is selected to continue the experiment, as shown in Figure 10. Figure 11 is the result after blur enhancement, the segmentation effects are shown in Figure 12 and Figure 13 respectively, and Figure 14 is the histogram of the original image.

The analysis shows that the image segmented by the FSO-ITS algorithm ideally separates the target area and the background area of the original image, and can fully reflect the edge contour of the image.

Next, the superiority of the proposed method is objectively analyzed by using image segmentation threshold, peak signal-to-noise ratio (PSNR) and information entropy and other evaluation criteria. Among them, the peak signal-to-noise ratio is based on the statistical and average calculation of the image pixel gray value, which is a commonly used indicator to measure signal distortion. Generally speaking, the larger the PSNR, the better the image quality. The formula for PSNR given below is expressed as follows:

where MSE is the mean squared error between the pre-encoded and decoded images. Table 1 below gives the comparison of the experimental results of the peak signal-to-noise ratio of the three algorithms.

Table 1 Comparison of PSNR experimental results of the three algorithms

As known from Table 1, the PSNR of the FSO-ITS algorithm proposed by the present invention is the largest, indicating that the distortion of the algorithm is the smallest, and the original information of the image is preserved most effectively. The following is an analysis of the information entropy of the experimental results. Image information entropy is a statistical form of features. It reflects the amount of information contained in the image. The larger the information entropy value of the segmented image, the greater the amount of information obtained from the source image, the richer the details of the segmented image, and the better the overall effect of the segmented image. The formula of information entropy H(x) is expressed as follows:

Table 2 Comparison of information entropy experimental results of the three algorithms

As can be seen from Table 2, the information entropy of the FSO-ITS algorithm proposed by the present invention is the largest, which means that the algorithm obtains the largest amount of information from the source image, and the edge of the segmented image is more complete, effectively improving the accuracy of image segmentation.

From the experimental results in Table 1 and Table 2, it can be seen that compared with the traditional Otsu algorithm, FSO-ITS has a better segmentation effect, continuous and complete edges, and better detail processing; compared with the traditional Otsu algorithm and Renyi entropy algorithm REA, FSO-ITS The peak signal-to-noise ratio and information entropy are the largest, the image distortion is the smallest, and the segmentation accuracy is the highest.

The specific implementations involved in the present invention have been provided above, but the present invention is not limited to the described implementations, such as using other forms of fuzzy membership functions, or the setting of specific parameters, the technical solution formed in this way is to carry out the above-mentioned embodiment Formed by fine-tuning, this technical solution still falls within the protection scope of the present invention.

Claims (10)

1. an image threshold segmentation method based on fuzzy sets and Otsu, is characterized in that, comprises the following steps: Use the fuzzy algorithm to perform fuzzy enhancement processing on the original image; Normalize the enhanced new image; Calculate the threshold th1 of Renyi entropy and the threshold th2 of Otsu respectively; Respectively find the inter-class variance corresponding to the threshold th1 and the inter-class variance corresponding to the threshold th2; Calculate the weight S 1 of the inter-class variance and the weight S 2 corresponding to the Renyi entropy according to the inter-class variance corresponding to the threshold _th1 and the threshold _th2 ; Obtain the segmentation threshold th=S ₁ th2+S ₂ th1; perform image segmentation according to the segmentation threshold th. 2. a kind of image threshold value segmentation method based on fuzzy set and Otsu according to claim 1, is characterized in that, the formula that obtains the threshold value th1 of Renyi entropy and the threshold value th2 of Otsu comprises: Among them, th1 is the threshold of Renyi entropy; th2 is the threshold of Otsu; E _O is the Renyi entropy of the target domain of the original image; E _B is the Renyi entropy of the background domain of the original image; L is the gray level of the original image; t is the gray level Threshold, and the value of t is [0, L–1]; w ₀ is the proportion of the first gray scale, w ₁ is the proportion of the second gray scale, the first gray scale and the second gray scale The grayscale is divided according to the grayscale threshold t; _σ0 is the grayscale mean square error of the image corresponding to the first type of grayscale, and _σ1 is the grayscale mean square error of the image corresponding to the second type of grayscale; The average gray value after the blur enhancement and the normalization processing. 3. a kind of image threshold segmentation method based on fuzzy set and Otsu according to claim 1 and 2, is characterized in that, the process of obtaining variance between described classes comprises: The between-class variances _σ2 and _σ3 of the threshold th1 of Renyi entropy are: The between-class variances _σ0 and _σ1 of Otsu's threshold th2 are: Among them, p _i is the probability of occurrence of the gray level with the number of pixels n _i ; u _T is the average gray level of the original image, L1=0, L2=L–1. 4. a kind of image threshold segmentation method based on fuzzy set and Otsu according to claim 3, is characterized in that, calculates the process that obtains the weight of described interclass variance comprising: Wherein, S ₁ is the weight of the inter-class variance, and S ₂ =1−S ₁ . 5. a kind of image threshold segmentation method based on fuzzy set and Otsu according to claim 4, is characterized in that, described normalization process comprises: Use min or max operator for edge extraction; Truncating the extracted edge data; The truncation process is: Wherein, T _r (u _ij ) is the edge data; u _ij is the membership function in the fuzzy algorithm. 6. a kind of image threshold segmentation device based on fuzzy set and Otsu, it is characterized in that: comprise processor and memory, described processor is stored with the instruction that realizes following method: Use the fuzzy algorithm to perform fuzzy enhancement processing on the original image; Normalize the enhanced new image; Calculate the threshold th1 of Renyi entropy and the threshold th2 of Otsu respectively; Respectively find the inter-class variance corresponding to the threshold th1 and the inter-class variance corresponding to the threshold th2; Calculate the weight S 1 of the inter-class variance and the weight S 2 corresponding to the Renyi entropy according to the inter-class variance corresponding to the threshold _th1 and the threshold _th2 ; Obtain the segmentation threshold th=S ₁ th2+S ₂ th1; perform image segmentation according to the segmentation threshold th. 7. A kind of image threshold segmentation device based on fuzzy set and Otsu according to claim 6, is characterized in that, the formula for obtaining the threshold th1 of Renyi entropy and the threshold th2 of Otsu comprises: Among them, th1 is the threshold of Renyi entropy; th2 is the threshold of Otsu; E _O is the Renyi entropy of the target domain of the original image; E _B is the Renyi entropy of the background domain of the original image; L is the gray level of the original image; t is the gray level Threshold, and the value of t is [0, L–1]; w ₀ is the proportion of the first gray scale, w ₁ is the proportion of the second gray scale, the first gray scale and the second gray scale The grayscale is divided according to the grayscale threshold t; _σ0 is the grayscale mean square error of the image corresponding to the first type of grayscale, and _σ1 is the grayscale mean square error of the image corresponding to the second type of grayscale; The average gray value after the blur enhancement and the normalization processing. 8. a kind of image threshold segmentation device based on fuzzy set and Otsu according to claim 6 or 7, is characterized in that, the process of obtaining variance between described classes comprises: The between-class variances _σ2 and _σ3 of the threshold th1 of Renyi entropy are: The between-class variances _σ0 and _σ1 of Otsu's threshold th2 are: Among them, p _i is the probability of occurrence of the gray level with the number of pixels n _i ; u _T is the average gray level of the original image, L1=0, L2=L–1. 9. a kind of image threshold segmentation device based on fuzzy set and Otsu according to claim 8, is characterized in that, calculates the process that obtains the weight of described interclass variance comprising: Wherein, S ₁ is the weight of the inter-class variance, and S ₂ =1−S ₁ . 10. a kind of image threshold segmentation device based on fuzzy set and Otsu according to claim 9, is characterized in that, described normalization process comprises: Use min or max operator for edge extraction; Truncating the extracted edge data; The truncation process is: Wherein, T _r (u _ij ) is the edge data; u _ij is the membership function in the fuzzy algorithm.